Abstract As the spatial resolution of satellite sensors increase, estimates of surface solar irradiance ( I SFC) from space borne observations of top of the atmosphere reflected radiance ( R TOA) can actually become less accurate because of enhanced three-dimensional (3-D) radiative effects that are not generally considered in most retrieval algorithms. An elementary approach to improving retrievals is to incorporate some form of spatial averaging through the use of superpixels. Determining the best approach for averaging and the optimum size of the superpixel is the objective of this study. It is achieved by examining the bias, correlation coefficient, and root-mean-square error (RMSE) between the true and estimated I SFC for a set of simulated cloud scenes and their radiative fields. Because of nonlinear effects, averaging within the superpixel should only be performed after I SFC is retrieved from R TOA independently for each pixel . For the combined set of stratocumulus and convective cloud fields that have a cloud fraction ranging from 0.25 to 1.0 used in this study, the optimal superpixel size is about 25 km. It is found that the bias is a function of solar and viewing angles and a correction is provided that improves the bias while leaving the correlation coefficient and root-mean-square error unchanged.